Race Condition Affecting vllm package, versions [,0.26.0)
Threat Intelligence
The probability is the direct output of the EPSS model, and conveys an overall sense of the threat of exploitation in the wild. The percentile measures the EPSS probability relative to all known EPSS scores. Note: This data is updated daily, relying on the latest available EPSS model version. Check out the EPSS documentation for more details.
Do your applications use this vulnerable package?
In a few clicks we can analyze your entire application and see what components are vulnerable in your application, and suggest you quick fixes.
Test your applications- Snyk IDSNYK-PYTHON-VLLM-18857125
- published16 Aug 2026
- disclosed13 Aug 2026
- credithexcraft-labs
Introduced: 13 Aug 2026
NewCVE-2026-73557 (opens in a new tab)How to fix?
Upgrade vllm to version 0.26.0 or higher.
Overview
vllm is an A high-throughput and memory-efficient inference and serving engine for LLMs
Affected versions of this package are vulnerable to Race Condition through the safe_load_prompt_embeds function in vllm/renderers/embed_utils.py when concurrent prompt parts are submitted to the /v1/chat/completions endpoint. An attacker can cause invalid sparse tensors to bypass intended safeguards and reach tensor.to_dense by exploiting race conditions in the global state management of torch.sparse.check_sparse_tensor_invariants.
Note: This is only exploitable if enable_prompt_embeds is enabled.